fix

Author: orca-zhang

src/llama.cpp

@@ -606,7 +606,7 @@ static struct ggml_tensor * llm_build_kqv(
         if (n_embd_head_v < n_embd_head_k) {
             cur = ggml_reshape_2d(ctx, ggml_cont(ctx, cur), n_embd_head_v_out*n_head, n_tokens);
             cur = ggml_cont(ctx, ggml_view_2d(ctx, ggml_cont(ctx, cur), n_embd_head_v*n_head, n_tokens,
-                               ggml_element_size(cur) * n_embd_head_v_out,
+                               ggml_element_size(cur) * n_embd_head_v_out*n_head,
                                0));
         } else {
             cur = ggml_reshape_2d(ctx, cur, n_embd_head_v*n_head, n_tokens);